Key optimization issues for the design of fuzzy-inference system

grantor: University of TorontoThe growing trend towards singleton chip fuzzy-inference systems poses the challenge and requirement of increasing information and decreasing the size of fuzzification, rule base, reasoning and defuzzification function blocks, without sacrificing inference rate and resolution. To meet this challenge, in this thesis we propose a new design for an algorithmic fuzzy-inference system, using digital memorization of operational results. This approach is taken to complement the more conventional arithmetic inference approach. This thesis proposes a new fuzzy-inference algorithm called the "Interval-Valued Representation and Reasoning method" (IVRR), and two optimization algorithms called the "Least Rule Base" and the "Operational Simplification of the Defuzzification". The Least Rule Base shows a redundant rule base that can be greatly reduced without any loss of information for the modified IVRR system, while in turn saving considerable chip area. The Operational Simplification of Defuzzification solves the problem of the operation complexity. With these three algorithms the design and simulation of binary fuzzy-inference systems are explored. A high performance digital fuzzy-inference system based on the Interval-Valued Representation and Reasoning method of fuzzy-inference can be developed with a capability of digital memorization, to record operational results while applying the Least Rule Base and the Operational Simplification of the Defuzzification algorithms. This fuzzy-inference method and the optimization algorithms are applied to an Inventory Capacity Planning System (ICPS) [Tûrks¸en and Berg, 1991]. The percentage of rule reduction is about 82%, the reduction of memorization bits of defuzzification is about 84% and the reduction of chip area is about 91.4% for the ICPS.Ph.D